https://hgpu.org/?p=4987
Programming finite-difference time-domain for graphics processor units using compute unified device architecture